On Mon, Aug 6, 2018 at 1:31 AM Christoph Hellwig <hch@xxxxxx> wrote: > > As our dear leader didn't like the ->poll_mask method this tries to > implement the behavior using plain old ->poll which is rather painful. I'm not seeing what's painful for this. Looking at the patches, this is *much* more straightforward than your previous patch, It adds refcounting to aio_iocb, but that's *much* better than messing up every other subsystem. Or is there some follow-up patches that are pending but you didn't post that are the painful part? Because the diffstat says that this second version is *way* less painful, at about 200 lines of code in a couple of files, mostly aio, vs ~700 lines of changes all over the place, together with a performance regression. Linus